DeepSeek V4 Pro 0813 (max) vs GLM-5
Head-to-head on the Towards AI writing benchmark: same 10 real YouTube scripts, five runs each, scored blind by a three-family judge panel. DeepSeek V4 Pro 0813 (max) leads overall, 84.1 to 81.9.
The verdict
DeepSeek V4 Pro 0813 (max) takes this matchup cleanly. It sits at #39 with an elo of 1847.7 against GLM-5 at #55 and 1690.6, a gap of 157.1 points, and the confidence intervals do not overlap. The overall scores are much closer: 84.09 versus 81.91, a spread of 2.18, and GLM-5 is the steadier performer with the lower deviation of the two. The separation comes from the slop family. DeepSeek posts 83.59 on anti-slop numbers where GLM-5 drops to 75.17, and it leads on slop, 88.75 to 85.04. GLM-5 answers with better cue quality, 81.97 to 79.38, and a hair more substance accuracy. Cost is a wash: both land around two cents per script, with DeepSeek a shade cheaper. Both ship open weights, so licensing settles nothing. On this evidence, DeepSeek is the default pick unless cue discipline matters most to you.
Pick DeepSeek V4 Pro 0813 (max) if you want the clearly stronger head-to-head record and much cleaner numbers handling, with slop control and hook strength both a step above GLM-5 at essentially the same price.
Pick GLM-5 if you value tighter run-to-run consistency and better cue quality, and can accept a lower board position in exchange.
Metric by metric
Blue bars: DeepSeek V4 Pro 0813 (max). Orange bars: GLM-5. Same 0–100 scale; the bold bar wins that metric.
Everything else that differs
| DeepSeek V4 Pro 0813 (max) | GLM-5 | |
|---|---|---|
| Overall / 100 | 84.1 | 81.9 |
| Writing Elo | 1848 | 1691 |
| Run-to-run spread (± overall std) | 6.490 | 3.590 |
| Cost per script (USD) | 0.022 | 0.024 |
| Avg latency (s) | 227.3 | 121.2 |
| Open weights | Yes | Yes |
Full scorecards: DeepSeek V4 Pro 0813 (max) · GLM-5. How scoring works: methodology.
← All comparisons